Log and transaction cleanups, factoring and bug fixes

index 4e4ee9a..639d965 100644
@@ -357,9 +357,8 @@ xfs_qm_dquot_logitem_init(
 	xfs_dq_logitem_t  *lp;
 	lp = &dqp->q_logitem;
 
-	lp->qli_item.li_type = XFS_LI_DQUOT;
-	lp->qli_item.li_ops = &xfs_dquot_item_ops;
-	lp->qli_item.li_mountp = dqp->q_mount;
+	xfs_log_item_init(dqp->q_mount, &lp->qli_item, XFS_LI_DQUOT,
+					&xfs_dquot_item_ops);
 	lp->qli_dquot = dqp;
 	lp->qli_format.qlf_type = XFS_LI_DQUOT;
 	lp->qli_format.qlf_id = be32_to_cpu(dqp->q_core.d_id);

@@ -586,11 +585,8 @@ xfs_qm_qoff_logitem_init(
 
 	qf = (xfs_qoff_logitem_t*) kmem_zalloc(sizeof(xfs_qoff_logitem_t), KM_SLEEP);
 
-	qf->qql_item.li_type = XFS_LI_QUOTAOFF;
-	if (start)
-		qf->qql_item.li_ops = &xfs_qm_qoffend_logitem_ops;
-	else
-		qf->qql_item.li_ops = &xfs_qm_qoff_logitem_ops;
+	xfs_log_item_init(mp, &qf->qql_item, XFS_LI_QUOTAOFF, start ?
+			&xfs_qm_qoffend_logitem_ops : &xfs_qm_qoff_logitem_ops);
 	qf->qql_item.li_mountp = mp;
 	qf->qql_format.qf_type = XFS_LI_QUOTAOFF;
 	qf->qql_format.qf_flags = flags;

index f3c49e6..aace237 100644
@@ -733,10 +733,7 @@ xfs_buf_item_init(
 
 	bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
 						    KM_SLEEP);
-	bip->bli_item.li_type = XFS_LI_BUF;
-	bip->bli_item.li_ops = &xfs_buf_item_ops;
-	bip->bli_item.li_mountp = mp;
-	bip->bli_item.li_ailp = mp->m_ail;
+	xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
 	bip->bli_buf = bp;
 	xfs_buf_hold(bp);
 	bip->bli_format.blf_type = XFS_LI_BUF;

index 6f35ed1..e461e93 100644
@@ -259,10 +259,7 @@ xfs_efi_init(xfs_mount_t	*mp,
 							     KM_SLEEP);
 	}
 
-	efip->efi_item.li_type = XFS_LI_EFI;
-	efip->efi_item.li_ops = &xfs_efi_item_ops;
-	efip->efi_item.li_mountp = mp;
-	efip->efi_item.li_ailp = mp->m_ail;
+	xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
 	efip->efi_format.efi_nextents = nextents;
 	efip->efi_format.efi_id = (__psint_t)(void*)efip;
 

@@ -554,10 +551,7 @@ xfs_efd_init(xfs_mount_t	*mp,
 							     KM_SLEEP);
 	}
 
-	efdp->efd_item.li_type = XFS_LI_EFD;
-	efdp->efd_item.li_ops = &xfs_efd_item_ops;
-	efdp->efd_item.li_mountp = mp;
-	efdp->efd_item.li_ailp = mp->m_ail;
+	xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
 	efdp->efd_efip = efip;
 	efdp->efd_format.efd_nextents = nextents;
 	efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;

index 7bfea85..32e4188 100644
@@ -865,17 +865,9 @@ xfs_inode_item_init(
 	ASSERT(ip->i_itemp == NULL);
 	iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
 
-	iip->ili_item.li_type = XFS_LI_INODE;
-	iip->ili_item.li_ops = &xfs_inode_item_ops;
-	iip->ili_item.li_mountp = mp;
-	iip->ili_item.li_ailp = mp->m_ail;
 	iip->ili_inode = ip;
-
-	/*
-	   We have zeroed memory. No need ...
-	   iip->ili_extents_buf = NULL;
-	 */
-
+	xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
+						&xfs_inode_item_ops);
 	iip->ili_format.ilf_type = XFS_LI_INODE;
 	iip->ili_format.ilf_ino = ip->i_ino;
 	iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;

index e8fba92..1f26a97 100644
@@ -648,6 +648,19 @@ xfs_log_unmount(xfs_mount_t *mp)
 	xlog_dealloc_log(mp->m_log);
 }
 
+void
+xfs_log_item_init(
+	struct xfs_mount	*mp,
+	struct xfs_log_item	*item,
+	int			type,
+	struct xfs_item_ops	*ops)
+{
+	item->li_mountp = mp;
+	item->li_ailp = mp->m_ail;
+	item->li_type = type;
+	item->li_ops = ops;
+}
+
 /*
  * Write region vectors to log.  The write happens using the space reservation
  * of the ticket (tic).  It is not a requirement that all writes for a given

index 97a24c7..f3a564d 100644
@@ -126,6 +126,13 @@ typedef struct xfs_log_callback {
 struct xfs_mount;
 struct xlog_in_core;
 struct xlog_ticket;
+struct xfs_log_item;
+struct xfs_item_ops;
+
+void	xfs_log_item_init(struct xfs_mount	*mp,
+			struct xfs_log_item	*item,
+			int			type,
+			struct xfs_item_ops	*ops);
 
 xfs_lsn_t xfs_log_done(struct xfs_mount *mp,
 		       struct xlog_ticket *ticket,

index fcaa62f..6537185 100644
@@ -562,18 +562,21 @@ DECLARE_EVENT_CLASS(xfs_inode_class,
 		__field(dev_t, dev)
 		__field(xfs_ino_t, ino)
 		__field(int, count)
+		__field(int, pincount)
 		__field(unsigned long, caller_ip)
 	),
 	TP_fast_assign(
 		__entry->dev = VFS_I(ip)->i_sb->s_dev;
 		__entry->ino = ip->i_ino;
 		__entry->count = atomic_read(&VFS_I(ip)->i_count);
+		__entry->pincount = atomic_read(&ip->i_pincount);
 		__entry->caller_ip = caller_ip;
 	),
-	TP_printk("dev %d:%d ino 0x%llx count %d caller %pf",
+	TP_printk("dev %d:%d ino 0x%llx count %d pincount %d caller %pf",
 		  MAJOR(__entry->dev), MINOR(__entry->dev),
 		  __entry->ino,
 		  __entry->count,
+		  __entry->pincount,
 		  (char *)__entry->caller_ip)
 )
 

@@ -583,6 +586,10 @@ DEFINE_EVENT(xfs_inode_class, name, \
 	TP_ARGS(ip, caller_ip))
 DEFINE_INODE_EVENT(xfs_ihold);
 DEFINE_INODE_EVENT(xfs_irele);
+DEFINE_INODE_EVENT(xfs_inode_pin);
+DEFINE_INODE_EVENT(xfs_inode_unpin);
+DEFINE_INODE_EVENT(xfs_inode_unpin_nowait);
+
 /* the old xfs_itrace_entry tracer - to be replaced by s.th. in the VFS */
 DEFINE_INODE_EVENT(xfs_inode);
 #define xfs_itrace_entry(ip)    \

index 0ffd564..8cd6e8d 100644
@@ -2449,6 +2449,8 @@ xfs_iunpin_nowait(
 {
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
 
+	trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
+
 	/* Give the log a push to start the unpinning I/O */
 	xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
 

index 32e4188..0347175 100644
@@ -543,6 +543,7 @@ xfs_inode_item_pin(
 {
 	ASSERT(xfs_isilocked(iip->ili_inode, XFS_ILOCK_EXCL));
 
+	trace_xfs_inode_pin(iip->ili_inode, _RET_IP_);
 	atomic_inc(&iip->ili_inode->i_pincount);
 }
 

@@ -561,6 +562,7 @@ xfs_inode_item_unpin(
 {
 	struct xfs_inode	*ip = iip->ili_inode;
 
+	trace_xfs_inode_unpin(ip, _RET_IP_);
 	ASSERT(atomic_read(&ip->i_pincount) > 0);
 	if (atomic_dec_and_test(&ip->i_pincount))
 		wake_up(&ip->i_ipin_wait);

index 639d965..165bbdf 100644
@@ -107,8 +107,7 @@ xfs_qm_dquot_logitem_pin(
 /* ARGSUSED */
 STATIC void
 xfs_qm_dquot_logitem_unpin(
-	xfs_dq_logitem_t *logitem,
-	int		  stale)
+	xfs_dq_logitem_t *logitem)
 {
 	xfs_dquot_t *dqp = logitem->qli_dquot;
 

@@ -123,7 +122,7 @@ xfs_qm_dquot_logitem_unpin_remove(
 	xfs_dq_logitem_t *logitem,
 	xfs_trans_t	 *tp)
 {
-	xfs_qm_dquot_logitem_unpin(logitem, 0);
+	xfs_qm_dquot_logitem_unpin(logitem);
 }
 
 /*

@@ -329,8 +328,7 @@ static struct xfs_item_ops xfs_dquot_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_qm_dquot_logitem_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*, int))
-					xfs_qm_dquot_logitem_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
 					xfs_qm_dquot_logitem_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))

@@ -425,7 +423,7 @@ xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t *qf)
  */
 /*ARGSUSED*/
 STATIC void
-xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t *qf, int stale)
+xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t *qf)
 {
 	return;
 }

@@ -536,8 +534,7 @@ static struct xfs_item_ops xfs_qm_qoffend_logitem_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_qm_qoff_logitem_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t* ,int))
-					xfs_qm_qoff_logitem_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
 					xfs_qm_qoff_logitem_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,

@@ -558,8 +555,7 @@ static struct xfs_item_ops xfs_qm_qoff_logitem_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_qm_qoff_logitem_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*, int))
-					xfs_qm_qoff_logitem_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
 					xfs_qm_qoff_logitem_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,

81 lines not on this page · raw message

index e461e93..409fe81 100644
@@ -106,7 +106,7 @@ xfs_efi_item_pin(xfs_efi_log_item_t *efip)
  */
 /*ARGSUSED*/
 STATIC void
-xfs_efi_item_unpin(xfs_efi_log_item_t *efip, int stale)
+xfs_efi_item_unpin(xfs_efi_log_item_t *efip)
 {
 	struct xfs_ail		*ailp = efip->efi_item.li_ailp;
 

@@ -224,7 +224,7 @@ static struct xfs_item_ops xfs_efi_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_efi_item_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_efi_item_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_efi_item_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_efi_item_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
 					xfs_efi_item_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_efi_item_trylock,

@@ -425,7 +425,7 @@ xfs_efd_item_pin(xfs_efd_log_item_t *efdp)
  */
 /*ARGSUSED*/
 STATIC void
-xfs_efd_item_unpin(xfs_efd_log_item_t *efdp, int stale)
+xfs_efd_item_unpin(xfs_efd_log_item_t *efdp)
 {
 	return;
 }

@@ -515,7 +515,7 @@ static struct xfs_item_ops xfs_efd_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_efd_item_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_efd_item_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_efd_item_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_efd_item_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
 					xfs_efd_item_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_efd_item_trylock,

index 0347175..cf8249a 100644
@@ -557,8 +557,7 @@ xfs_inode_item_pin(
 /* ARGSUSED */
 STATIC void
 xfs_inode_item_unpin(
-	xfs_inode_log_item_t	*iip,
-	int			stale)
+	xfs_inode_log_item_t	*iip)
 {
 	struct xfs_inode	*ip = iip->ili_inode;
 

@@ -574,7 +573,7 @@ xfs_inode_item_unpin_remove(
 	xfs_inode_log_item_t	*iip,
 	xfs_trans_t		*tp)
 {
-	xfs_inode_item_unpin(iip, 0);
+	xfs_inode_item_unpin(iip);
 }
 
 /*

@@ -840,7 +839,7 @@ static struct xfs_item_ops xfs_inode_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_inode_item_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_inode_item_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_inode_item_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_inode_item_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
 					xfs_inode_item_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_inode_item_trylock,

index f73e358..6962f2b 100644
@@ -1451,6 +1451,6 @@ xfs_trans_chunk_committed(
 		 * flags, if anyone else stales the buffer we do not
 		 * want to pay any attention to it.
 		 */
-		IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);
+		IOP_UNPIN(lip);
 	}
 }

index 79c8bab..82574ef 100644
@@ -159,7 +159,6 @@ typedef struct xfs_log_item_desc {
 
 #define XFS_LID_DIRTY		0x1
 #define XFS_LID_PINNED		0x2
-#define XFS_LID_BUF_STALE	0x8
 
 /*
  * This structure is used to maintain a chunk list of log_item_desc

@@ -833,7 +832,7 @@ typedef struct xfs_item_ops {
 	uint (*iop_size)(xfs_log_item_t *);
 	void (*iop_format)(xfs_log_item_t *, struct xfs_log_iovec *);
 	void (*iop_pin)(xfs_log_item_t *);
-	void (*iop_unpin)(xfs_log_item_t *, int);
+	void (*iop_unpin)(xfs_log_item_t *);
 	void (*iop_unpin_remove)(xfs_log_item_t *, struct xfs_trans *);
 	uint (*iop_trylock)(xfs_log_item_t *);
 	void (*iop_unlock)(xfs_log_item_t *);

@@ -846,7 +845,7 @@ typedef struct xfs_item_ops {
 #define IOP_SIZE(ip)		(*(ip)->li_ops->iop_size)(ip)
 #define IOP_FORMAT(ip,vp)	(*(ip)->li_ops->iop_format)(ip, vp)
 #define IOP_PIN(ip)		(*(ip)->li_ops->iop_pin)(ip)
-#define IOP_UNPIN(ip, flags)	(*(ip)->li_ops->iop_unpin)(ip, flags)
+#define IOP_UNPIN(ip)		(*(ip)->li_ops->iop_unpin)(ip)
 #define IOP_UNPIN_REMOVE(ip,tp) (*(ip)->li_ops->iop_unpin_remove)(ip, tp)
 #define IOP_TRYLOCK(ip)		(*(ip)->li_ops->iop_trylock)(ip)
 #define IOP_UNLOCK(ip)		(*(ip)->li_ops->iop_unlock)(ip)

index fb58636..4e1b689 100644
@@ -696,7 +696,6 @@ xfs_trans_log_buf(xfs_trans_t	*tp,
 
 	tp->t_flags |= XFS_TRANS_DIRTY;
 	lidp->lid_flags |= XFS_LID_DIRTY;
-	lidp->lid_flags &= ~XFS_LID_BUF_STALE;
 	bip->bli_flags |= XFS_BLI_LOGGED;
 	xfs_buf_item_log(bip, first, last);
 }

@@ -782,7 +781,7 @@ xfs_trans_binval(
 	bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
 	memset((char *)(bip->bli_format.blf_data_map), 0,
 	      (bip->bli_format.blf_map_size * sizeof(uint)));
-	lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
+	lidp->lid_flags |= XFS_LID_DIRTY;
 	tp->t_flags |= XFS_TRANS_DIRTY;
 }
 

index 95a4813..528a3b4 100644
@@ -3834,7 +3834,7 @@ xfs_bmap_add_attrfork(
 	}
 	if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
 		goto error2;
-	error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES);
+	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 	ASSERT(ip->i_df.if_ext_max ==
 	       XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
 	return error;

index 6962f2b..ac9af40 100644
@@ -47,8 +47,6 @@
 
 
 STATIC void	xfs_trans_apply_sb_deltas(xfs_trans_t *);
-STATIC uint	xfs_trans_count_vecs(xfs_trans_t *);
-STATIC void	xfs_trans_fill_vecs(xfs_trans_t *, xfs_log_iovec_t *);
 STATIC void	xfs_trans_uncommit(xfs_trans_t *, uint);
 STATIC void	xfs_trans_committed(xfs_trans_t *, int);
 STATIC void	xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int);

@@ -764,94 +762,126 @@ xfs_trans_unreserve_and_mod_sb(
 	}
 }
 
-
 /*
- * xfs_trans_commit
- *
- * Commit the given transaction to the log a/synchronously.
- *
- * XFS disk error handling mechanism is not based on a typical
- * transaction abort mechanism. Logically after the filesystem
- * gets marked 'SHUTDOWN', we can't let any new transactions
- * be durable - ie. committed to disk - because some metadata might
- * be inconsistent. In such cases, this returns an error, and the
- * caller may assume that all locked objects joined to the transaction
- * have already been unlocked as if the commit had succeeded.
- * Do not reference the transaction structure after this call.
+ * Total up the number of log iovecs needed to commit this
+ * transaction.  The transaction itself needs one for the
+ * transaction header.  Ask each dirty item in turn how many
+ * it needs to get the total.
  */
- /*ARGSUSED*/
-int
-_xfs_trans_commit(
-	xfs_trans_t	*tp,
-	uint		flags,
-	int		*log_flushed)
+static uint
+xfs_trans_count_vecs(
+	xfs_trans_t	*tp)
 {
-	xfs_log_iovec_t		*log_vector;
-	int			nvec;
-	xfs_mount_t		*mp;
-	xfs_lsn_t		commit_lsn;
-	/* REFERENCED */
-	int			error;
-	int			log_flags;
-	int			sync;
-#define	XFS_TRANS_LOGVEC_COUNT	16
-	xfs_log_iovec_t		log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
-	struct xlog_in_core	*commit_iclog;
-	int			shutdown;
+	int			nvecs;
+	xfs_log_item_desc_t	*lidp;
 
-	commit_lsn = -1;
+	nvecs = 1;
+	lidp = xfs_trans_first_item(tp);
+	ASSERT(lidp != NULL);
 
-	/*
-	 * Determine whether this commit is releasing a permanent
-	 * log reservation or not.
+	/* In the non-debug case we need to start bailing out if we
+	 * didn't find a log_item here, return zero and let trans_commit
+	 * deal with it.
 	 */
-	if (flags & XFS_TRANS_RELEASE_LOG_RES) {
-		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
-		log_flags = XFS_LOG_REL_PERM_RESERV;
-	} else {
-		log_flags = 0;
+	if (lidp == NULL)
+		return 0;
+
+	while (lidp != NULL) {
+		/*
+		 * Skip items which aren't dirty in this transaction.
+		 */
+		if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
+			lidp = xfs_trans_next_item(tp, lidp);
+			continue;
+		}
+		lidp->lid_size = IOP_SIZE(lidp->lid_item);
+		nvecs += lidp->lid_size;
+		lidp = xfs_trans_next_item(tp, lidp);
 	}
-	mp = tp->t_mountp;
+
+	return nvecs;
+}
+
+/*
+ * Fill in the vector with pointers to data to be logged
+ * by this transaction.  The transaction header takes
+ * the first vector, and then each dirty item takes the
+ * number of vectors it indicated it needed in xfs_trans_count_vecs().
+ *
+ * As each item fills in the entries it needs, also pin the item
+ * so that it cannot be flushed out until the log write completes.
+ */
+static void
+xfs_trans_fill_vecs(
+	struct xfs_trans	*tp,
+	struct xfs_log_iovec	*log_vector)
+{
+	xfs_log_item_desc_t	*lidp;
+	struct xfs_log_iovec	*vecp;
+	uint			nitems;
 
 	/*
-	 * If there is nothing to be logged by the transaction,
-	 * then unlock all of the items associated with the
-	 * transaction and free the transaction structure.
-	 * Also make sure to return any reserved blocks to
-	 * the free pool.
+	 * Skip over the entry for the transaction header, we'll
+	 * fill that in at the end.
 	 */
-shut_us_down:
-	shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
-	if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
-		xfs_trans_unreserve_and_mod_sb(tp);
+	vecp = log_vector + 1;
+
+	nitems = 0;
+	lidp = xfs_trans_first_item(tp);
+	ASSERT(lidp);
+	while (lidp) {
+		/* Skip items which aren't dirty in this transaction. */
+		if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
+			lidp = xfs_trans_next_item(tp, lidp);
+			continue;
+		}
+
 		/*
-		 * It is indeed possible for the transaction to be
-		 * not dirty but the dqinfo portion to be. All that
-		 * means is that we have some (non-persistent) quota
-		 * reservations that need to be unreserved.
+		 * The item may be marked dirty but not log anything.  This can
+		 * be used to get called when a transaction is committed.
 		 */
-		xfs_trans_unreserve_and_mod_dquots(tp);
-		if (tp->t_ticket) {
-			commit_lsn = xfs_log_done(mp, tp->t_ticket,
-							NULL, log_flags);
-			if (commit_lsn == -1 && !shutdown)
-				shutdown = XFS_ERROR(EIO);
-		}
-		current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
-		xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
-		xfs_trans_free_busy(tp);
-		xfs_trans_free(tp);
-		XFS_STATS_INC(xs_trans_empty);
-		return (shutdown);
+		if (lidp->lid_size)
+			nitems++;
+		IOP_FORMAT(lidp->lid_item, vecp);
+		vecp += lidp->lid_size;
+		IOP_PIN(lidp->lid_item);
+		lidp = xfs_trans_next_item(tp, lidp);
 	}
-	ASSERT(tp->t_ticket != NULL);
 
 	/*
-	 * If we need to update the superblock, then do it now.
+	 * Now that we've counted the number of items in this transaction, fill
+	 * in the transaction header. Note that the transaction header does not
+	 * have a log item.
 	 */
-	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
-		xfs_trans_apply_sb_deltas(tp);
-	xfs_trans_apply_dquot_deltas(tp);
+	tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
+	tp->t_header.th_type = tp->t_type;
+	tp->t_header.th_num_items = nitems;
+	log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
+	log_vector->i_len = sizeof(xfs_trans_header_t);
+	log_vector->i_type = XLOG_REG_TYPE_TRANSHDR;
+}
+
+/*
+ * Format the transaction direct to the iclog. This isolates the physical
+ * transaction commit operation from the logical operation and hence allows
+ * other methods to be introduced without affecting the existing commit path.
+ */
+static int
+xfs_trans_commit_iclog(
+	struct xfs_mount	*mp,
+	struct xfs_trans	*tp,
+	xfs_lsn_t		*commit_lsn,
+	int			flags)
+{
+	int			shutdown;
+	int			error;
+	int			log_flags = 0;
+	struct xlog_in_core	*commit_iclog;
+#define XFS_TRANS_LOGVEC_COUNT  16
+	struct xfs_log_iovec	log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
+	struct xfs_log_iovec	*log_vector;
+	uint			nvec;
+
 
 	/*
 	 * Ask each log item how many log_vector entries it will

@@ -861,8 +891,7 @@ shut_us_down:
 	 */
 	nvec = xfs_trans_count_vecs(tp);
 	if (nvec == 0) {
-		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
-		goto shut_us_down;
+		return ENOMEM;	/* triggers a shutdown! */
 	} else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
 		log_vector = log_vector_fast;
 	} else {

@@ -877,6 +906,9 @@ shut_us_down:
 	 */
 	xfs_trans_fill_vecs(tp, log_vector);
 
+	if (flags & XFS_TRANS_RELEASE_LOG_RES)
+		log_flags = XFS_LOG_REL_PERM_RESERV;
+
 	error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
 
 	/*

@@ -884,18 +916,17 @@ shut_us_down:
 	 * at any time after this call.  However, all the items associated
 	 * with the transaction are still locked and pinned in memory.
 	 */
-	commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
+	*commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
 
-	tp->t_commit_lsn = commit_lsn;
-	if (nvec > XFS_TRANS_LOGVEC_COUNT) {
-		kmem_free(log_vector);
-	}
+	tp->t_commit_lsn = *commit_lsn;
+        if (nvec > XFS_TRANS_LOGVEC_COUNT)
+                kmem_free(log_vector);
 
 	/*
 	 * If we got a log write error. Unpin the logitems that we
 	 * had pinned, clean up, free trans structure, and return error.
 	 */
-	if (error || commit_lsn == -1) {
+	if (error || *commit_lsn == -1) {
 		current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
 		xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
 		return XFS_ERROR(EIO);

@@ -909,8 +940,6 @@ shut_us_down:
 	 */
 	xfs_trans_unreserve_and_mod_sb(tp);
 
-	sync = tp->t_flags & XFS_TRANS_SYNC;
-
 	/*
 	 * Tell the LM to call the transaction completion routine
 	 * when the log write with LSN commit_lsn completes (e.g.

@@ -953,7 +982,7 @@ shut_us_down:
 	 * the commit lsn of this transaction for dependency tracking
 	 * purposes.
 	 */
-	xfs_trans_unlock_items(tp, commit_lsn);
+	xfs_trans_unlock_items(tp, *commit_lsn);
 
 	/*
 	 * If we detected a log error earlier, finish committing

@@ -973,7 +1002,92 @@ shut_us_down:
 	 * and the items are released we can finally allow the iclog to
 	 * go to disk.
 	 */
-	error = xfs_log_release_iclog(mp, commit_iclog);
+	return xfs_log_release_iclog(mp, commit_iclog);
+}
+
+
+/*
+ * xfs_trans_commit
+ *
+ * Commit the given transaction to the log a/synchronously.
+ *
+ * XFS disk error handling mechanism is not based on a typical
+ * transaction abort mechanism. Logically after the filesystem
+ * gets marked 'SHUTDOWN', we can't let any new transactions
+ * be durable - ie. committed to disk - because some metadata might
+ * be inconsistent. In such cases, this returns an error, and the
+ * caller may assume that all locked objects joined to the transaction
+ * have already been unlocked as if the commit had succeeded.
+ * Do not reference the transaction structure after this call.
+ */
+ /*ARGSUSED*/
+int
+_xfs_trans_commit(
+	xfs_trans_t	*tp,
+	uint		flags,
+	int		*log_flushed)
+{
+	xfs_mount_t		*mp = tp->t_mountp;
+	xfs_lsn_t		commit_lsn = -1;
+	int			error;
+	int			log_flags = 0;
+	int			sync = tp->t_flags & XFS_TRANS_SYNC;
+	int			shutdown;
+
+	/*
+	 * Determine whether this commit is releasing a permanent
+	 * log reservation or not.
+	 */
+	if (flags & XFS_TRANS_RELEASE_LOG_RES) {
+		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
+		log_flags = XFS_LOG_REL_PERM_RESERV;
+	}
+
+	/*
+	 * If there is nothing to be logged by the transaction,
+	 * then unlock all of the items associated with the
+	 * transaction and free the transaction structure.
+	 * Also make sure to return any reserved blocks to
+	 * the free pool.
+	 */
+shut_us_down:
+	shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
+	if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
+		xfs_trans_unreserve_and_mod_sb(tp);
+		/*
+		 * It is indeed possible for the transaction to be
+		 * not dirty but the dqinfo portion to be. All that
+		 * means is that we have some (non-persistent) quota
+		 * reservations that need to be unreserved.
+		 */
+		xfs_trans_unreserve_and_mod_dquots(tp);
+		if (tp->t_ticket) {
+			commit_lsn = xfs_log_done(mp, tp->t_ticket,
+							NULL, log_flags);
+			if (commit_lsn == -1 && !shutdown)
+				shutdown = XFS_ERROR(EIO);
+		}
+		current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
+		xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
+		xfs_trans_free_busy(tp);
+		xfs_trans_free(tp);
+		XFS_STATS_INC(xs_trans_empty);
+		return (shutdown);
+	}
+	ASSERT(tp->t_ticket != NULL);
+
+	/*
+	 * If we need to update the superblock, then do it now.
+	 */
+	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
+		xfs_trans_apply_sb_deltas(tp);
+	xfs_trans_apply_dquot_deltas(tp);
+
+	error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags);
+	if (error == ENOMEM) {
+		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
+		goto shut_us_down;
+	}
 
 	/*
 	 * If the transaction needs to be synchronous, then force the

@@ -992,47 +1106,6 @@ shut_us_down:
 	return (error);
 }
 
-
-/*
- * Total up the number of log iovecs needed to commit this
- * transaction.  The transaction itself needs one for the
- * transaction header.  Ask each dirty item in turn how many
- * it needs to get the total.
- */
-STATIC uint
-xfs_trans_count_vecs(
-	xfs_trans_t	*tp)
-{
-	int			nvecs;
-	xfs_log_item_desc_t	*lidp;
-
-	nvecs = 1;
-	lidp = xfs_trans_first_item(tp);
-	ASSERT(lidp != NULL);
-
-	/* In the non-debug case we need to start bailing out if we
-	 * didn't find a log_item here, return zero and let trans_commit
-	 * deal with it.
-	 */
-	if (lidp == NULL)
-		return 0;
-
-	while (lidp != NULL) {
-		/*
-		 * Skip items which aren't dirty in this transaction.
-		 */
-		if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
-			lidp = xfs_trans_next_item(tp, lidp);
-			continue;
-		}
-		lidp->lid_size = IOP_SIZE(lidp->lid_item);
-		nvecs += lidp->lid_size;
-		lidp = xfs_trans_next_item(tp, lidp);
-	}
-
-	return nvecs;
-}
-
 /*
  * Called from the trans_commit code when we notice that
  * the filesystem is in the middle of a forced shutdown.

@@ -1063,68 +1136,6 @@ xfs_trans_uncommit(
 }
 
 /*
- * Fill in the vector with pointers to data to be logged
- * by this transaction.  The transaction header takes
- * the first vector, and then each dirty item takes the
- * number of vectors it indicated it needed in xfs_trans_count_vecs().
- *
- * As each item fills in the entries it needs, also pin the item
- * so that it cannot be flushed out until the log write completes.
- */
-STATIC void
-xfs_trans_fill_vecs(
-	xfs_trans_t		*tp,
-	xfs_log_iovec_t		*log_vector)
-{
-	xfs_log_item_desc_t	*lidp;
-	xfs_log_iovec_t		*vecp;
-	uint			nitems;
-
-	/*
-	 * Skip over the entry for the transaction header, we'll
-	 * fill that in at the end.
-	 */
-	vecp = log_vector + 1;		/* pointer arithmetic */
-
-	nitems = 0;
-	lidp = xfs_trans_first_item(tp);
-	ASSERT(lidp != NULL);
-	while (lidp != NULL) {
-		/*
-		 * Skip items which aren't dirty in this transaction.
-		 */
-		if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
-			lidp = xfs_trans_next_item(tp, lidp);
-			continue;
-		}
-		/*
-		 * The item may be marked dirty but not log anything.
-		 * This can be used to get called when a transaction
-		 * is committed.
-		 */
-		if (lidp->lid_size) {
-			nitems++;
-		}
-		IOP_FORMAT(lidp->lid_item, vecp);
-		vecp += lidp->lid_size;		/* pointer arithmetic */
-		IOP_PIN(lidp->lid_item);
-		lidp = xfs_trans_next_item(tp, lidp);
-	}
-
-	/*
-	 * Now that we've counted the number of items in this
-	 * transaction, fill in the transaction header.
-	 */
-	tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
-	tp->t_header.th_type = tp->t_type;
-	tp->t_header.th_num_items = nitems;
-	log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
-	log_vector->i_len = sizeof(xfs_trans_header_t);
-	log_vector->i_type = XLOG_REG_TYPE_TRANSHDR;
-}
-
-
-/*
  * Unlock all of the transaction's items and free the transaction.
  * The transaction must not have modified any of its items, because
  * there is no way to restore them to their previous state.

index ac9af40..abc9874 100644
@@ -49,7 +49,6 @@
 STATIC void	xfs_trans_apply_sb_deltas(xfs_trans_t *);
 STATIC void	xfs_trans_uncommit(xfs_trans_t *, uint);
 STATIC void	xfs_trans_committed(xfs_trans_t *, int);
-STATIC void	xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int);
 STATIC void	xfs_trans_free(xfs_trans_t *);
 
 kmem_zone_t	*xfs_trans_zone;

@@ -1296,60 +1295,86 @@ xfs_trans_roll(
 }
 
 /*
- * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
+ * The committed item processing consists of calling the committed routine of
+ * each logged item, updating the item's position in the AIL if necessary, and
+ * unpinning each item.  If the committed routine returns -1, then do nothing
+ * further with the item because it may have been freed.
  *
- * This is typically called by the LM when a transaction has been fully
- * committed to disk.  It needs to unpin the items which have
- * been logged by the transaction and update their positions
- * in the AIL if necessary.
- * This also gets called when the transactions didn't get written out
- * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
+ * Since items are unlocked when they are copied to the incore log, it is
+ * possible for two transactions to be completing and manipulating the same
+ * item simultaneously.  The AIL lock will protect the lsn field of each item.
+ * The value of this field can never go backwards.
  *
- * Call xfs_trans_chunk_committed() to process the items in
- * each chunk.
+ * We unpin the items after repositioning them in the AIL, because otherwise
+ * they could be immediately flushed and we'd have to race with the flusher
+ * trying to pull the item from the AIL as we add it.
  */
-STATIC void
-xfs_trans_committed(
-	xfs_trans_t	*tp,
-	int		abortflag)
+static void
+xfs_trans_item_committed(
+	xfs_log_item_t	*lip,
+	xfs_lsn_t	commit_lsn,
+	int		aborted)
 {
-	xfs_log_item_chunk_t	*licp;
-	xfs_log_item_chunk_t	*next_licp;
-	xfs_log_busy_chunk_t	*lbcp;
-	xfs_log_busy_slot_t	*lbsp;
-	int			i;
+	xfs_lsn_t	item_lsn;
+	struct xfs_ail	*ailp;
+
+	if (aborted)
+		lip->li_flags |= XFS_LI_ABORTED;
 
 	/*
-	 * Call the transaction's completion callback if there
-	 * is one.
+	 * Send in the ABORTED flag to the COMMITTED routine so that it knows
+	 * whether the transaction was aborted or not.
 	 */
-	if (tp->t_callback != NULL) {
-		tp->t_callback(tp, tp->t_callarg);
-	}
+	item_lsn = IOP_COMMITTED(lip, commit_lsn);
 
 	/*
-	 * Special case the chunk embedded in the transaction.
+	 * If the committed routine returns -1, item has been freed.
 	 */
-	licp = &(tp->t_items);
-	if (!(xfs_lic_are_all_free(licp))) {
-		xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
-	}
+	if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
+		return;
 
 	/*
-	 * Process the items in each chunk in turn.
+	 * If the returned lsn is greater than what it contained before, update
+	 * the location of the item in the AIL.  If it is not, then do nothing.
+	 * Items can never move backwards in the AIL.
+	 *
+	 * While the new lsn should usually be greater, it is possible that a
+	 * later transaction completing simultaneously with an earlier one
+	 * using the same item could complete first with a higher lsn.  This
+	 * would cause the earlier transaction to fail the test below.
 	 */
-	licp = licp->lic_next;
-	while (licp != NULL) {
-		ASSERT(!xfs_lic_are_all_free(licp));
-		xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
-		next_licp = licp->lic_next;
-		kmem_free(licp);
-		licp = next_licp;
+	ailp = lip->li_ailp;
+	spin_lock(&ailp->xa_lock);
+	if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
+		/*
+		 * This will set the item's lsn to item_lsn and update the
+		 * position of the item in the AIL.
+		 *
+		 * xfs_trans_ail_update() drops the AIL lock.
+		 */
+		xfs_trans_ail_update(ailp, lip, item_lsn);
+	} else {
+		spin_unlock(&ailp->xa_lock);
 	}
 
 	/*
-	 * Clear all the per-AG busy list items listed in this transaction
+	 * Now that we've repositioned the item in the AIL, unpin it so it can
+	 * be flushed. Pass information about buffer stale state down from the
+	 * log item flags, if anyone else stales the buffer we do not want to
+	 * pay any attention to it.
 	 */
+	IOP_UNPIN(lip);
+}
+
+/* Clear all the per-AG busy list items listed in this transaction */
+static void
+xfs_trans_clear_busy_extents(
+	struct xfs_trans	*tp)
+{
+	xfs_log_busy_chunk_t	*lbcp;
+	xfs_log_busy_slot_t	*lbsp;
+	int			i;
+
 	lbcp = &tp->t_busy;
 	while (lbcp != NULL) {
 		for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {

@@ -1361,107 +1386,49 @@ xfs_trans_committed(
 		lbcp = lbcp->lbc_next;
 	}
 	xfs_trans_free_busy(tp);
-
-	/*
-	 * That's it for the transaction structure.  Free it.
-	 */
-	xfs_trans_free(tp);
 }
 
 /*
- * This is called to perform the commit processing for each
- * item described by the given chunk.
- *
- * The commit processing consists of unlocking items which were
- * held locked with the SYNC_UNLOCK attribute, calling the committed
- * routine of each logged item, updating the item's position in the AIL
- * if necessary, and unpinning each item.  If the committed routine
- * returns -1, then do nothing further with the item because it
- * may have been freed.
- *
- * Since items are unlocked when they are copied to the incore
- * log, it is possible for two transactions to be completing
- * and manipulating the same item simultaneously.  The AIL lock
- * will protect the lsn field of each item.  The value of this
- * field can never go backwards.
+ * This is typically called by the LM when a transaction has been fully
+ * committed to disk.  It needs to unpin the items which have
+ * been logged by the transaction and update their positions
+ * in the AIL if necessary.
  *
- * We unpin the items after repositioning them in the AIL, because
- * otherwise they could be immediately flushed and we'd have to race
- * with the flusher trying to pull the item from the AIL as we add it.
+ * This also gets called when the transactions didn't get written out
+ * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
  */
 STATIC void
-xfs_trans_chunk_committed(
-	xfs_log_item_chunk_t	*licp,
-	xfs_lsn_t		lsn,
-	int			aborted)
+xfs_trans_committed(
+	xfs_trans_t	*tp,
+	int		abortflag)
 {
 	xfs_log_item_desc_t	*lidp;
-	xfs_log_item_t		*lip;
-	xfs_lsn_t		item_lsn;
-	int			i;
-
-	lidp = licp->lic_descs;
-	for (i = 0; i < licp->lic_unused; i++, lidp++) {
-		struct xfs_ail		*ailp;
-
-		if (xfs_lic_isfree(licp, i)) {
-			continue;
-		}
-
-		lip = lidp->lid_item;
-		if (aborted)
-			lip->li_flags |= XFS_LI_ABORTED;
-
-		/*
-		 * Send in the ABORTED flag to the COMMITTED routine
-		 * so that it knows whether the transaction was aborted
-		 * or not.
-		 */
-		item_lsn = IOP_COMMITTED(lip, lsn);
+	xfs_log_item_chunk_t	*licp;
+	xfs_log_item_chunk_t	*next_licp;
 
-		/*
-		 * If the committed routine returns -1, make
-		 * no more references to the item.
-		 */
-		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
-			continue;
-		}
+	/*
+	 * Call the transaction's completion callback if there
+	 * is one.
+	 */
+	if (tp->t_callback != NULL) {
+		tp->t_callback(tp, tp->t_callarg);
+	}
 
-		/*
-		 * If the returned lsn is greater than what it
-		 * contained before, update the location of the
-		 * item in the AIL.  If it is not, then do nothing.
-		 * Items can never move backwards in the AIL.
-		 *
-		 * While the new lsn should usually be greater, it
-		 * is possible that a later transaction completing
-		 * simultaneously with an earlier one using the
-		 * same item could complete first with a higher lsn.
-		 * This would cause the earlier transaction to fail
-		 * the test below.
-		 */
-		ailp = lip->li_ailp;
-		spin_lock(&ailp->xa_lock);
-		if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
-			/*
-			 * This will set the item's lsn to item_lsn
-			 * and update the position of the item in
-			 * the AIL.
-			 *
-			 * xfs_trans_ail_update() drops the AIL lock.
-			 */
-			xfs_trans_ail_update(ailp, lip, item_lsn);
-		} else {
-			spin_unlock(&ailp->xa_lock);
-		}
+	for (lidp = xfs_trans_first_item(tp);
+	     lidp != NULL;
+	     lidp = xfs_trans_next_item(tp, lidp)) {
+		xfs_trans_item_committed(lidp->lid_item, tp->t_lsn, abortflag);
+	}
 
-		/*
-		 * Now that we've repositioned the item in the AIL,
-		 * unpin it so it can be flushed. Pass information
-		 * about buffer stale state down from the log item
-		 * flags, if anyone else stales the buffer we do not
-		 * want to pay any attention to it.
-		 */
-		IOP_UNPIN(lip);
+	/* free the item chunks, ignoring the embedded chunk */
+	licp = tp->t_items.lic_next;
+	while (licp != NULL) {
+		next_licp = licp->lic_next;
+		ASSERT(xfs_lic_are_all_free(licp));
+		kmem_free(licp);
+		licp = next_licp;
 	}
+
+	xfs_trans_clear_busy_extents(tp);
+	xfs_trans_free(tp);
 }

index 1f26a97..7c6b0cd 100644
@@ -664,7 +664,10 @@ xfs_log_item_init(
 /*
  * Write region vectors to log.  The write happens using the space reservation
  * of the ticket (tic).  It is not a requirement that all writes for a given
- * transaction occur with one call to xfs_log_write().
+ * transaction occur with one call to xfs_log_write(). However, it is important
+ * to note that the transaction reservation code makes an assumption about the
+ * number of log headers a transaction requires that may be violated if you
+ * don't pass all the transaction vectors in one call....
  */
 int
 xfs_log_write(

@@ -3156,14 +3159,16 @@ xfs_log_ticket_get(
  * Allocate and initialise a new log ticket.
  */
 STATIC xlog_ticket_t *
-xlog_ticket_alloc(xlog_t		*log,
-		int		unit_bytes,
-		int		cnt,
-		char		client,
-		uint		xflags)
+xlog_ticket_alloc(
+	struct log	*log,
+	int		unit_bytes,
+	int		cnt,
+	char		client,
+	uint		xflags)
 {
-	xlog_ticket_t	*tic;
+	struct xlog_ticket *tic;
 	uint		num_headers;
+	int		iclog_space;
 
 	tic = kmem_zone_zalloc(xfs_log_ticket_zone, KM_SLEEP|KM_MAYFAIL);
 	if (!tic)

@@ -3207,16 +3212,40 @@ xlog_ticket_alloc(xlog_t		*log,
 	/* for start-rec */
 	unit_bytes += sizeof(xlog_op_header_t);
 
-	/* for LR headers */
-	num_headers = ((unit_bytes + log->l_iclog_size-1) >> log->l_iclog_size_log);
+	/*
+	 * for LR headers - the space for data in an iclog is the size minus
+	 * the space used for the headers. If we use the iclog size, then we
+	 * undercalculate the number of headers required.
+	 *
+	 * Furthermore - the addition of op headers for split-recs might
+	 * increase the space required enough to require more log and op
+	 * headers, so take that into account too.
+	 *
+	 * IMPORTANT: This reservation makes the assumption that if this
+	 * transaction is the first in an iclog and hence has the LR headers
+	 * accounted to it, then the remaining space in the iclog is
+	 * exclusively for this transaction.  i.e. if the transaction is larger
+	 * than the iclog, it will be the only thing in that iclog.
+	 * Fundamentally, this means we must pass the entire log vector to
+	 * xlog_write to guarantee this.
+	 */
+	iclog_space = log->l_iclog_size - log->l_iclog_hsize;
+	num_headers = (unit_bytes + iclog_space - 1) / iclog_space;
+
+	/* for split-recs - ophdrs added when data split over LRs */
+	unit_bytes += sizeof(xlog_op_header_t) * num_headers;
+
+	/* add extra header reservations if we overrun */
+	while (!num_headers ||
+	       ((unit_bytes + iclog_space - 1) / iclog_space) > num_headers) {
+		unit_bytes += sizeof(xlog_op_header_t);
+		num_headers++;
+	}
 	unit_bytes += log->l_iclog_hsize * num_headers;
 
 	/* for commit-rec LR header - note: padding will subsume the ophdr */
 	unit_bytes += log->l_iclog_hsize;
 
-	/* for split-recs - ophdrs added when data split over LRs */
-	unit_bytes += sizeof(xlog_op_header_t) * num_headers;
-
 	/* for roundoff padding for transaction data and one for commit record */
 	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
 	    log->l_mp->m_sb.sb_logsunit > 1) {

@@ -3238,7 +3267,7 @@ xlog_ticket_alloc(xlog_t		*log,
 	tic->t_trans_type	= 0;
 	if (xflags & XFS_LOG_PERM_RESERV)
 		tic->t_flags |= XLOG_TIC_PERM_RESERV;
-	sv_init(&(tic->t_wait), SV_DEFAULT, "logtick");
+	sv_init(&tic->t_wait, SV_DEFAULT, "logtick");
 
 	xlog_tic_reset_res(tic);
 

index 7c6b0cd..afb27cf 100644
@@ -50,7 +50,7 @@ kmem_zone_t	*xfs_log_ticket_zone;
 	  (off) += (bytes);}
 
 /* Local miscellaneous function prototypes */
-STATIC int	 xlog_commit_record(xfs_mount_t *mp, xlog_ticket_t *ticket,
+STATIC int	 xlog_commit_record(struct log *log, struct xlog_ticket *ticket,
 				    xlog_in_core_t **, xfs_lsn_t *);
 STATIC xlog_t *  xlog_alloc_log(xfs_mount_t	*mp,
 				xfs_buftarg_t	*log_target,

@@ -59,11 +59,9 @@ STATIC xlog_t *  xlog_alloc_log(xfs_mount_t	*mp,
 STATIC int	 xlog_space_left(xlog_t *log, int cycle, int bytes);
 STATIC int	 xlog_sync(xlog_t *log, xlog_in_core_t *iclog);
 STATIC void	 xlog_dealloc_log(xlog_t *log);
-STATIC int	 xlog_write(xfs_mount_t *mp, xfs_log_iovec_t region[],
-			    int nentries, struct xlog_ticket *tic,
-			    xfs_lsn_t *start_lsn,
-			    xlog_in_core_t **commit_iclog,
-			    uint flags);
+STATIC int	 xlog_write(struct log *log, struct xfs_log_vec *log_vector,
+			    struct xlog_ticket *tic, xfs_lsn_t *start_lsn,
+			    xlog_in_core_t **commit_iclog, uint flags);
 
 /* local state machine functions */
 STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);

@@ -258,7 +256,7 @@ xfs_log_done(
 	     * If we get an error, just continue and give back the log ticket.
 	     */
 	    (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
-	     (xlog_commit_record(mp, ticket, iclog, &lsn)))) {
+	     (xlog_commit_record(log, ticket, iclog, &lsn)))) {
 		lsn = (xfs_lsn_t) -1;
 		if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
 			flags |= XFS_LOG_REL_PERM_RESERV;

@@ -516,18 +514,10 @@ xfs_log_unmount_write(xfs_mount_t *mp)
 #ifdef DEBUG
 	xlog_in_core_t	 *first_iclog;
 #endif
-	xfs_log_iovec_t  reg[1];
 	xlog_ticket_t	*tic = NULL;
 	xfs_lsn_t	 lsn;
 	int		 error;
 
-	/* the data section must be 32 bit size aligned */
-	struct {
-	    __uint16_t magic;
-	    __uint16_t pad1;
-	    __uint32_t pad2; /* may as well make it 64 bits */
-	} magic = { XLOG_UNMOUNT_TYPE, 0, 0 };
-
 	/*
 	 * Don't write out unmount record on read-only mounts.
 	 * Or, if we are doing a forced umount (typically because of IO errors).

@@ -549,16 +539,31 @@ xfs_log_unmount_write(xfs_mount_t *mp)
 	} while (iclog != first_iclog);
 #endif
 	if (! (XLOG_FORCED_SHUTDOWN(log))) {
-		reg[0].i_addr = (void*)&magic;
-		reg[0].i_len  = sizeof(magic);
-		reg[0].i_type = XLOG_REG_TYPE_UNMOUNT;
-
 		error = xfs_log_reserve(mp, 600, 1, &tic,
 					XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
 		if (!error) {
+			/* the data section must be 32 bit size aligned */
+			struct {
+			    __uint16_t magic;
+			    __uint16_t pad1;
+			    __uint32_t pad2; /* may as well make it 64 bits */
+			} magic = {
+				.magic = XLOG_UNMOUNT_TYPE,
+			};
+			struct xfs_log_iovec reg = {
+				.i_addr = (void *)&magic,
+				.i_len = sizeof(magic),
+				.i_type = XLOG_REG_TYPE_UNMOUNT,
+			};
+			struct xfs_log_vec vec = {
+				.lv_niovecs = 1,
+			};
+			/* sigh. c99 initialisers don't work on anon unions */
+			vec.lv_iovecp = ®
+
 			/* remove inited flag */
-			((xlog_ticket_t *)tic)->t_flags = 0;
-			error = xlog_write(mp, reg, 1, tic, &lsn,
+			tic->t_flags = 0;
+			error = xlog_write(log, &vec, tic, &lsn,
 					   NULL, XLOG_UNMOUNT_TRANS);
 			/*
 			 * At this point, we're umounting anyway,

@@ -679,11 +684,16 @@ xfs_log_write(
 {
 	struct log		*log = mp->m_log;
 	int			error;
+	struct xfs_log_vec	vec = {
+		.lv_niovecs = nentries,
+	};
+	/* sigh. c99 initialisers don't work on anon unions */
+	vec.lv_iovecp = reg;
 
 	if (XLOG_FORCED_SHUTDOWN(log))
 		return XFS_ERROR(EIO);
 
-	error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0);
+	error = xlog_write(log, &vec, tic, start_lsn, NULL, 0);
 	if (error)
 		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
 	return error;

@@ -1176,26 +1186,32 @@ out:
  * ticket.  Return the lsn of the commit record.
  */
 STATIC int
-xlog_commit_record(xfs_mount_t  *mp,
-		   xlog_ticket_t *ticket,
-		   xlog_in_core_t **iclog,
-		   xfs_lsn_t	*commitlsnp)
+xlog_commit_record(
+	struct log		*log,
+	struct xlog_ticket	*ticket,
+	struct xlog_in_core	**iclog,
+	xfs_lsn_t		*commitlsnp)
 {
-	int		error;
-	xfs_log_iovec_t	reg[1];
-
-	reg[0].i_addr = NULL;
-	reg[0].i_len = 0;
-	reg[0].i_type = XLOG_REG_TYPE_COMMIT;
+	struct xfs_mount *mp = log->l_mp;
+	int	error;
+	struct xfs_log_iovec reg = {
+		.i_addr = NULL,
+		.i_len = 0,
+		.i_type = XLOG_REG_TYPE_COMMIT,
+	};
+	struct xfs_log_vec vec = {
+		.lv_niovecs = 1,
+	};
+	/* sigh. c99 initialisers don't work on anon unions */
+	vec.lv_iovecp = ®
 
 	ASSERT_ALWAYS(iclog);
-	if ((error = xlog_write(mp, reg, 1, ticket, commitlsnp,
-			       iclog, XLOG_COMMIT_TRANS))) {
+	error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
+					XLOG_COMMIT_TRANS);
+	if (error)
 		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
-	}
 	return error;
-}	/* xlog_commit_record */
-
+}
 
 /*
  * Push on the buffer cache code if we ever use more than 75% of the on-disk

@@ -1657,15 +1673,13 @@ xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket)
  */
 STATIC int
 xlog_write(
-	struct xfs_mount	*mp,
-	struct xfs_log_iovec	reg[],
-	int			nentries,
+	struct log		*log,
+	struct xfs_log_vec	*log_vector,
 	struct xlog_ticket	*ticket,
 	xfs_lsn_t		*start_lsn,
 	struct xlog_in_core	**commit_iclog,
 	uint			flags)
 {
-    xlog_t	     *log = mp->m_log;
     xlog_in_core_t   *iclog = NULL;  /* ptr to current in-core log */
     xlog_op_header_t *logop_head;    /* ptr to log operation header */
     __psint_t	     ptr;	     /* copy address into data region */

@@ -1681,6 +1695,8 @@ xlog_write(
     int		     contwr;	     /* continued write of in-core log? */
     int		     error;
     int		     record_cnt = 0, data_cnt = 0;
+    struct xfs_log_iovec *reg = log_vector->lv_iovecp;
+    int		     nentries = log_vector->lv_niovecs;
 
     partial_copy_len = partial_copy = 0;
 

@@ -1702,16 +1718,16 @@ xlog_write(
     contwr = *start_lsn = 0;
 
     if (ticket->t_curr_res < len) {
-	xlog_print_tic_res(mp, ticket);
+	xlog_print_tic_res(log->l_mp, ticket);
 #ifdef DEBUG
 	xlog_panic(
 		"xfs_log_write: reservation ran out. Need to up reservation");
 #else
 	/* Customer configurable panic */
-	xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp,
+	xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, log->l_mp,
 		"xfs_log_write: reservation ran out. Need to up reservation");
 	/* If we did not panic, shutdown the filesystem */
-	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
+	xfs_force_shutdown(log->l_mp, SHUTDOWN_CORRUPT_INCORE);
 #endif
     } else
 	ticket->t_curr_res -= len;

@@ -1777,7 +1793,7 @@ xlog_write(
 	    case XFS_LOG:
 		break;
 	    default:
-		xfs_fs_cmn_err(CE_WARN, mp,
+		xfs_fs_cmn_err(CE_WARN, log->l_mp,
 		    "Bad XFS transaction clientid 0x%x in ticket 0x%p",
 		    logop_head->oh_clientid, ticket);
 		return XFS_ERROR(EIO);

index f3a564d..229d1f3 100644
@@ -110,6 +110,12 @@ typedef struct xfs_log_iovec {
 	uint		i_type;		/* type of region */
 } xfs_log_iovec_t;
 
+struct xfs_log_vec {
+	struct xfs_log_vec	*lv_next;	/* next lv in build list */
+	int			lv_niovecs;	/* number of iovecs in lv */
+	struct xfs_log_iovec	*lv_iovecp;	/* iovec array */
+};
+
 /*
  * Structure used to pass callback function and the function's argument
  * to the log manager.

index afb27cf..38b4dc4 100644
@@ -1632,6 +1632,193 @@ xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket)
 }
 
 /*
+ * Calculate the potential space needed by the log vector.  Each region gets
+ * its own xlog_op_header_t and may need to be double word aligned. If regions
+ * get split, we'll account for the extra headers then.
+ */
+static int
+xlog_write_calc_vec_length(
+	struct xlog_ticket *ticket,
+	struct xfs_log_vec *log_vector)
+{
+	struct xfs_log_vec *lv;
+	int		headers = 0;
+	int		len = 0;
+	int		index;
+
+	if (ticket->t_flags & XLOG_TIC_INITED)
+		headers++;
+
+	for (lv = log_vector; lv; lv = lv->lv_next) {
+		struct xfs_log_iovec	*vecp;
+
+		headers += lv->lv_niovecs;
+		vecp = lv->lv_iovecp;
+		for (index = 0; index < lv->lv_niovecs; index++) {
+			len += vecp->i_len;
+			xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
+			vecp++;
+		}
+	}
+
+	ticket->t_res_num_ophdrs += headers;
+	len += headers * sizeof(xlog_op_header_t);
+
+	return len;
+}
+
+/*
+ * If first write for transaction, insert start record.  We can't be trying to
+ * commit if we are inited.  We can't have any "partial_copy" if we are inited.
+ */
+static int
+xlog_write_start_rec(
+	__psint_t	ptr,
+	xlog_ticket_t	*ticket)
+{
+	xlog_op_header_t *logop_head;
+
+	if (!(ticket->t_flags & XLOG_TIC_INITED))
+		return 0;
+	logop_head = (xlog_op_header_t *)ptr;
+	logop_head->oh_tid = cpu_to_be32(ticket->t_tid);
+	logop_head->oh_clientid = ticket->t_clientid;
+	logop_head->oh_len = 0;
+	logop_head->oh_flags = XLOG_START_TRANS;
+	logop_head->oh_res2 = 0;
+	ticket->t_flags &= ~XLOG_TIC_INITED;
+	return sizeof(xlog_op_header_t);
+}
+
+static xlog_op_header_t *
+xlog_write_setup_ophdr(
+	struct log	*log,
+	__psint_t	ptr,
+	xlog_ticket_t	*ticket,
+	uint		flags)
+{
+	xlog_op_header_t *logop_head;
+
+	/* Copy log operation header directly into data section */
+	logop_head = (xlog_op_header_t *)ptr;
+	logop_head->oh_tid = cpu_to_be32(ticket->t_tid);
+	logop_head->oh_clientid = ticket->t_clientid;
+	logop_head->oh_res2 = 0;
+
+	/* are we copying a commit or unmount record? */
+	logop_head->oh_flags = flags;
+
+	/*
+	 * We've seen logs corrupted with bad transaction client ids.  This
+	 * makes sure that XFS doesn't generate them on.  Turn this into an EIO
+	 * and shut down the filesystem.
+	 */
+	switch (logop_head->oh_clientid)  {
+	case XFS_TRANSACTION:
+	case XFS_VOLUME:
+	case XFS_LOG:
+		break;
+	default:
+		xfs_fs_cmn_err(CE_WARN, log->l_mp,
+			"Bad XFS transaction clientid 0x%x in ticket 0x%p",
+			logop_head->oh_clientid, ticket);
+		return NULL;
+	}
+	return logop_head;
+}
+
+
+/*
+ * Set up the parameters of the region copy into the log. This has
+ * to handle region write split across multiple log buffers - this
+ * state is kept external to this function so that this code can
+ * can be written in an obvious, self documenting manner.
+ */
+static int
+xlog_write_setup_copy(
+	xlog_ticket_t	*ticket,
+	xlog_op_header_t *ophdr,
+	int		space_available,
+	int		space_required,
+	int		*copy_off,
+	int		*copy_len,
+	int		*last_was_partial_copy,
+	int		*bytes_consumed)
+{
+	int	still_to_copy;
+
+	still_to_copy = space_required - *bytes_consumed;
+	*copy_off = *bytes_consumed;
+
+	if (still_to_copy <= space_available) {
+		/* write of region completes here */
+		*copy_len = still_to_copy;
+		ophdr->oh_len = cpu_to_be32(*copy_len);
+		if (*last_was_partial_copy)
+			ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
+		*last_was_partial_copy = 0;
+		*bytes_consumed = 0;
+		return 0;
+	}
+
+	/* partial write of region, needs extra log op header reservation */
+	*copy_len = space_available;
+	ophdr->oh_len = cpu_to_be32(*copy_len);
+	ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
+	if (*last_was_partial_copy)
+		ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
+	*bytes_consumed += *copy_len;
+	(*last_was_partial_copy)++;
+	ticket->t_curr_res -= sizeof(xlog_op_header_t);
+	ticket->t_res_num_ophdrs++;
+	return sizeof(xlog_op_header_t);
+}
+
+static int
+xlog_write_copy_finish(
+	xlog_t		*log,
+	xlog_in_core_t	*iclog,
+	uint		flags,
+	int		*record_cnt,
+	int		*data_cnt,
+	int		*partial_copy,
+	int		*partial_copy_len,
+	int		log_offset,
+	xlog_in_core_t	**commit_iclog)
+{
+	if (*partial_copy) {
+		/*
+		 * iclog has already been marked WANT_SYNC by
+		 * xlog_state_get_iclog_space
+		 */
+		xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
+		*record_cnt = 0;
+		*data_cnt = 0;
+		return xlog_state_release_iclog(log, iclog);
+	}
+
+	*partial_copy = 0;
+	*partial_copy_len = 0;
+
+	if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
+		/* no more space in this iclog - push it. */
+		xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
+		*record_cnt = 0;
+		*data_cnt = 0;
+
+		spin_lock(&log->l_icloglock);
+		xlog_state_want_sync(log, iclog);
+		spin_unlock(&log->l_icloglock);
+
+		if (!commit_iclog)
+			return xlog_state_release_iclog(log, iclog);
+		ASSERT(flags & XLOG_COMMIT_TRANS);
+		*commit_iclog = iclog;
+	}
+	return 0;
+}
+
+/*
  * Write some region out to in-core log
  *
  * This will be called when writing externally provided regions or when

@@ -1680,203 +1867,149 @@ xlog_write(
 	struct xlog_in_core	**commit_iclog,
 	uint			flags)
 {
-    xlog_in_core_t   *iclog = NULL;  /* ptr to current in-core log */
-    xlog_op_header_t *logop_head;    /* ptr to log operation header */
-    __psint_t	     ptr;	     /* copy address into data region */
-    int		     len;	     /* # xlog_write() bytes 2 still copy */
-    int		     index;	     /* region index currently copying */
-    int		     log_offset;     /* offset (from 0) into data region */
-    int		     start_rec_copy; /* # bytes to copy for start record */
-    int		     partial_copy;   /* did we split a region? */
-    int		     partial_copy_len;/* # bytes copied if split region */
-    int		     need_copy;	     /* # bytes need to memcpy this region */
-    int		     copy_len;	     /* # bytes actually memcpy'ing */
-    int		     copy_off;	     /* # bytes from entry start */
-    int		     contwr;	     /* continued write of in-core log? */
-    int		     error;
-    int		     record_cnt = 0, data_cnt = 0;
-    struct xfs_log_iovec *reg = log_vector->lv_iovecp;
-    int		     nentries = log_vector->lv_niovecs;
-
-    partial_copy_len = partial_copy = 0;
-
-    /* Calculate potential maximum space.  Each region gets its own
-     * xlog_op_header_t and may need to be double word aligned.
-     */
-    len = 0;
-    if (ticket->t_flags & XLOG_TIC_INITED) {    /* acct for start rec of xact */
-	len += sizeof(xlog_op_header_t);
-	ticket->t_res_num_ophdrs++;
-    }
-
-    for (index = 0; index < nentries; index++) {
-	len += sizeof(xlog_op_header_t);	    /* each region gets >= 1 */
-	ticket->t_res_num_ophdrs++;
-	len += reg[index].i_len;
-	xlog_tic_add_region(ticket, reg[index].i_len, reg[index].i_type);
-    }
-    contwr = *start_lsn = 0;
+	xlog_in_core_t	*iclog = NULL;  /* ptr to current in-core log */
+	xlog_op_header_t *logop_head;	/* ptr to log operation header */
+	struct xfs_log_iovec *vecp;
+	struct xfs_log_vec *lv;
+	__psint_t	ptr;		/* copy address into data region */
+	int		len = 0;	/* # xlog_write() bytes 2 still copy */
+	int		index;		/* region index currently copying */
+	int		log_offset;	/* offset (from 0) into data region */
+	int		start_rec_copy;	/* # bytes to copy for start record */
+	int		partial_copy;	/* did we split a region? */
+	int		partial_copy_len;/* # bytes copied if split region */
+	int		copy_len;	/* # bytes actually memcpy'ing */
+	int		copy_off;	/* # bytes from entry start */
+	int		contwr = 0;	/* continued write of in-core log? */
+	int		error = 0;
+	int		record_cnt = 0;
+	int		data_cnt = 0;
+
+	partial_copy_len = partial_copy = 0;
+	if (commit_iclog)
+		*commit_iclog = NULL;
+
+	len += xlog_write_calc_vec_length(ticket, log_vector);
 
-    if (ticket->t_curr_res < len) {
-	xlog_print_tic_res(log->l_mp, ticket);
+	ticket->t_curr_res -= len;
+	if (ticket->t_curr_res < 0) {
+		xlog_print_tic_res(log->l_mp, ticket);
 #ifdef DEBUG
-	xlog_panic(
-		"xfs_log_write: reservation ran out. Need to up reservation");
+		xlog_panic("xfs_log_write: "
+				"reservation ran out. Need to up reservation");
 #else
-	/* Customer configurable panic */
-	xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, log->l_mp,
-		"xfs_log_write: reservation ran out. Need to up reservation");
-	/* If we did not panic, shutdown the filesystem */
-	xfs_force_shutdown(log->l_mp, SHUTDOWN_CORRUPT_INCORE);
+		/* Customer configurable panic */
+		xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, log->l_mp,
+				"xfs_log_write: "
+				"reservation ran out. Need to up reservation");
+		/* If we did not panic, shutdown the filesystem */
+		xfs_force_shutdown(log->l_mp, SHUTDOWN_CORRUPT_INCORE);
 #endif
-    } else
-	ticket->t_curr_res -= len;
-
-    for (index = 0; index < nentries; ) {
-	if ((error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
-					       &contwr, &log_offset)))
-		return error;
-
-	ASSERT(log_offset <= iclog->ic_size - 1);
-	ptr = (__psint_t) ((char *)iclog->ic_datap+log_offset);
+	}
 
-	/* start_lsn is the first lsn written to. That's all we need. */
-	if (! *start_lsn)
-	    *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
+	index = 0;
+	lv = log_vector;
+	vecp = lv->lv_iovecp;
+	while (lv && index < lv->lv_niovecs) {
 
-	/* This loop writes out as many regions as can fit in the amount
-	 * of space which was allocated by xlog_state_get_iclog_space().
-	 */
-	while (index < nentries) {
-	    ASSERT(reg[index].i_len % sizeof(__int32_t) == 0);
-	    ASSERT((__psint_t)ptr % sizeof(__int32_t) == 0);
-	    start_rec_copy = 0;
-
-	    /* If first write for transaction, insert start record.
-	     * We can't be trying to commit if we are inited.  We can't
-	     * have any "partial_copy" if we are inited.
-	     */
-	    if (ticket->t_flags & XLOG_TIC_INITED) {
-		logop_head		= (xlog_op_header_t *)ptr;
-		logop_head->oh_tid	= cpu_to_be32(ticket->t_tid);
-		logop_head->oh_clientid = ticket->t_clientid;
-		logop_head->oh_len	= 0;
-		logop_head->oh_flags    = XLOG_START_TRANS;
-		logop_head->oh_res2	= 0;
-		ticket->t_flags		&= ~XLOG_TIC_INITED;	/* clear bit */
-		record_cnt++;
-
-		start_rec_copy = sizeof(xlog_op_header_t);
-		xlog_write_adv_cnt(ptr, len, log_offset, start_rec_copy);
-	    }
+		error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
+						       &contwr, &log_offset);
+		if (error)
+			return error;
 
-	    /* Copy log operation header directly into data section */
-	    logop_head			= (xlog_op_header_t *)ptr;
-	    logop_head->oh_tid		= cpu_to_be32(ticket->t_tid);
-	    logop_head->oh_clientid	= ticket->t_clientid;
-	    logop_head->oh_res2		= 0;
+		ASSERT(log_offset <= iclog->ic_size - 1);
+		ptr = (__psint_t) ((char *)iclog->ic_datap + log_offset);
 
-	    /* header copied directly */
-	    xlog_write_adv_cnt(ptr, len, log_offset, sizeof(xlog_op_header_t));
+		/* start_lsn is the first lsn written to. That's all we need. */
+		if (start_lsn)
+			*start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
 
-	    /* are we copying a commit or unmount record? */
-	    logop_head->oh_flags = flags;
+		/*
+		 * This loop writes out as many regions as can fit in the amount
+		 * of space which was allocated by xlog_state_get_iclog_space().
+		 */
+		while (lv && index < lv->lv_niovecs) {
+			struct xfs_log_iovec *reg = &vecp[index];
+
+			ASSERT(reg->i_len % sizeof(__int32_t) == 0);
+			ASSERT((__psint_t)ptr % sizeof(__int32_t) == 0);
+			start_rec_copy = 0;
+
+			start_rec_copy = xlog_write_start_rec(ptr, ticket);
+			if (start_rec_copy) {
+				record_cnt++;
+				xlog_write_adv_cnt(ptr, len, log_offset,
+							start_rec_copy);
+			}
 
-	    /*
-	     * We've seen logs corrupted with bad transaction client
-	     * ids.  This makes sure that XFS doesn't generate them on.
-	     * Turn this into an EIO and shut down the filesystem.
-	     */
-	    switch (logop_head->oh_clientid)  {
-	    case XFS_TRANSACTION:
-	    case XFS_VOLUME:
-	    case XFS_LOG:
-		break;
-	    default:
-		xfs_fs_cmn_err(CE_WARN, log->l_mp,
-		    "Bad XFS transaction clientid 0x%x in ticket 0x%p",
-		    logop_head->oh_clientid, ticket);
-		return XFS_ERROR(EIO);
-	    }
+			logop_head = xlog_write_setup_ophdr(log, ptr,
+							ticket, flags);
+			if (!logop_head)
+				return XFS_ERROR(EIO);
+			xlog_write_adv_cnt(ptr, len, log_offset,
+						sizeof(xlog_op_header_t));
+
+			len += xlog_write_setup_copy(ticket, logop_head,
+					iclog->ic_size - log_offset,
+					reg->i_len, &copy_off, &copy_len,
+					&partial_copy, &partial_copy_len);
+
+			xlog_verify_dest_ptr(log, ptr);
+
+			/* copy region */
+			ASSERT(copy_len >= 0);
+			memcpy((void *)ptr, reg->i_addr + copy_off, copy_len);
+			xlog_write_adv_cnt(ptr, len, log_offset, copy_len);
+
+			/* copy_len is total bytes copied, including headers */
+			copy_len += start_rec_copy + sizeof(xlog_op_header_t);
+			record_cnt++;
+			data_cnt += contwr ? copy_len : 0;
+
+			error = xlog_write_copy_finish(log, iclog, flags,
+					&record_cnt, &data_cnt,
+					&partial_copy, &partial_copy_len,
+					log_offset, commit_iclog);
+			if (error)
+				return error;
 
-	    /* Partial write last time? => (partial_copy != 0)
-	     * need_copy is the amount we'd like to copy if everything could
-	     * fit in the current memcpy.
-	     */
-	    need_copy =	reg[index].i_len - partial_copy_len;
-
-	    copy_off = partial_copy_len;
-	    if (need_copy <= iclog->ic_size - log_offset) { /*complete write */
-	        copy_len = need_copy;
-		logop_head->oh_len = cpu_to_be32(copy_len);
-		if (partial_copy)
-		    logop_head->oh_flags|= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
-		partial_copy_len = partial_copy = 0;
-	    } else {					    /* partial write */
-		copy_len = iclog->ic_size - log_offset;
-		logop_head->oh_len = cpu_to_be32(copy_len);
-		logop_head->oh_flags |= XLOG_CONTINUE_TRANS;
-		if (partial_copy)
-			logop_head->oh_flags |= XLOG_WAS_CONT_TRANS;
-		partial_copy_len += copy_len;
-		partial_copy++;
-		len += sizeof(xlog_op_header_t); /* from splitting of region */
-		/* account for new log op header */
-		ticket->t_curr_res -= sizeof(xlog_op_header_t);
-		ticket->t_res_num_ophdrs++;
-	    }
-	    xlog_verify_dest_ptr(log, ptr);
-
-	    /* copy region */
-	    ASSERT(copy_len >= 0);
-	    memcpy((xfs_caddr_t)ptr, reg[index].i_addr + copy_off, copy_len);
-	    xlog_write_adv_cnt(ptr, len, log_offset, copy_len);
-
-	    /* make copy_len total bytes copied, including headers */
-	    copy_len += start_rec_copy + sizeof(xlog_op_header_t);
-	    record_cnt++;
-	    data_cnt += contwr ? copy_len : 0;
-	    if (partial_copy) {			/* copied partial region */
-		    /* already marked WANT_SYNC by xlog_state_get_iclog_space */
-		    xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
-		    record_cnt = data_cnt = 0;
-		    if ((error = xlog_state_release_iclog(log, iclog)))
-			    return error;
-		    break;			/* don't increment index */
-	    } else {				/* copied entire region */
-		index++;
-		partial_copy_len = partial_copy = 0;
-
-		if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
-		    xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
-		    record_cnt = data_cnt = 0;
-		    spin_lock(&log->l_icloglock);
-		    xlog_state_want_sync(log, iclog);
-		    spin_unlock(&log->l_icloglock);
-		    if (commit_iclog) {
-			ASSERT(flags & XLOG_COMMIT_TRANS);
-			*commit_iclog = iclog;
-		    } else if ((error = xlog_state_release_iclog(log, iclog)))
-			   return error;
-		    if (index == nentries)
-			    return 0;		/* we are done */
-		    else
-			    break;
+			/*
+			 * if we had a partial copy, we need to get more iclog
+			 * space but we don't want to increment the region
+			 * index because there is still more is this region to
+			 * write.
+			 *
+			 * If we completed writing this region, and we flushed
+			 * the iclog (indicated by resetting of the record
+			 * count), then we also need to get more log space. If
+			 * this was the last record, though, we are done and
+			 * can just return.
+			 */
+			if (partial_copy)
+				break;
+			if (++index == lv->lv_niovecs) {
+				lv = lv->lv_next;
+				index = 0;
+				if (lv)
+					vecp = lv->lv_iovecp;
+			}
+			if (record_cnt == 0) {
+				if (!lv)
+					return 0; /* last region, done! */
+				break;
+			}
 		}
-	    } /* if (partial_copy) */
-	} /* while (index < nentries) */
-    } /* for (index = 0; index < nentries; ) */
-    ASSERT(len == 0);
-
-    xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
-    if (commit_iclog) {
-	ASSERT(flags & XLOG_COMMIT_TRANS);
-	*commit_iclog = iclog;
-	return 0;
-    }
-    return xlog_state_release_iclog(log, iclog);
-}	/* xlog_write */
+	}
+	ASSERT(len == 0);
 
+	xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
+	if (commit_iclog) {
+		ASSERT(flags & XLOG_COMMIT_TRANS);
+		*commit_iclog = iclog;
+		return 0;
+	}
+	return xlog_state_release_iclog(log, iclog);
+}
 
 /*****************************************************************************
  *